BUS PASSENGER DEMAND PREDICTION BASED ON TIME SERIES NEURAL NETWORK REGRESSION WITH EXIT-ONLY SMART CARD DATA

Estimation of bus passenger demand at station level is a vital component of public transportation management which can be used to enhance service quality, maximizing benefits for operators, minimizing waiting time for passengers and reduce congestion level at peak hours. In this paper, we used exit-only AFC transactions data and AVL transactions data of a typical feeder route of Tehran, Iran over a typical month (25th October 2015 to 24th November 2015). Since in exit-only AFC transactions there is no information about boarding and alighting station of passengers, it is a challenge to determine this information about bus passengers trip, and this challenges was done using integration between AFC data and AVL data. For prediction demand of bus passengers at station level in time intervals, Levenberg-Marquardt algorithm (LM) is an iterative technique was used to design neural network time series regression and neural network curve fit. As the results shows, neural network curve fit has 0.753 goodness-of-fit and neural network time series regression has 0.816 goodness-of-fit which supports that used data has time series variant and it has better performance in time series analysis. Suggested network can predict demand of bus passengers with high accuracy which leads operators toward optimize time-table of bus routes, vehicle scheduling and crew scheduling.